| Name | operagents JSON |
| Version |
0.0.15
JSON |
| download |
| home_page | None |
| Summary | Dynamic, highly customizable language agents framework |
| upload_time | 2024-05-11 10:15:14 |
| maintainer | None |
| docs_url | None |
| author | yanyongyu |
| requires_python | <4.0,>=3.10 |
| license | MIT |
| keywords |
agent
llm
langchain
sop
langgraph
|
| VCS |
|
| bugtrack_url |
|
| requirements |
No requirements were recorded.
|
| Travis-CI |
No Travis.
|
| coveralls test coverage |
No coveralls.
|
# Operagents
## Installation
Install the latest version with:
```bash
pip install operagents
# or use poetry
poetry add operagents
# or use pdm
pdm add operagents
# or use uv
uv pip install operagents
```
## Concepts
### Agent
An agent is a human or a language model that can act as [characters](#character) and use [props](#prop) in the opera scenes. The agent can communicate with others by observing and acting. Every agent has a backend (e.g. user, openai api) to generate the response and own memory to store the long-term / short-term information.
### Scene
A scene is a part of the opera that contains a number of [characters](#character). Every scene has a [flow](#flow) and a [director](#director) to control the whole [session](#session) process. The scene can also have a prepare section to do some initialization work before the scene starts.
### Character
A character is a role in the [scene](#scene). Every character has a name, a description, and a list of [props](#prop). When the scene starts, an agent will act as the character and communicate with others.
### Flow
The flow is used to control the order of the characters' acting in the [scene](#scene).
### Director
The director is used to decide whether to end the current scene and which scene to play next.
### Prop
A prop is a tool that can be used by the [agents](#agent) to improve their acting. Agents can get external information by using props.
### Timeline
The timeline is the main runtime component of the opera to manage the [session](#session) process. It runs the current session and switches between sessions. The timeline also records the global information of the opera, and can be shared by all agents.
### Session
A session indicates a single run of the scene. It contains an unique identifier and its corresponding scene.
## Usage
The common way to use operagents is to write a config file and run the opera with the `operagents` command-line tool.
### Start writing a config file
Create a `config.yaml` file with the following basic content:
```yaml
# yaml-language-server: $schema=https://operagents.yyydl.top/schemas/config.schema.json
agents:
opening_scene: ""
scenes:
```
The first line is a comment that tells the YAML Language Server to use the schema from the specified URL. This will enable autocompletion and validation in your editor.
The schema is related to the version of the operagents framework you are using. The URL is in the format `https://operagents.yyydl.top/schemas/config-<version>.schema.json`, where `<version>` is the version of the framework, e.g. `0.0.1`. If no version is specified, the latest (master) version will be used.
### The Template config
Before writing the agent and scene configs, we need to learn about the template config.
Operagents uses templates to generate the context input for the language model. A template is a string in [jinja](https://jinja.palletsprojects.com/) format. You can use jinja2 syntax with provided context varaibles to control the input to the language model.
A template config can be in the following format:
1. simple string template
```yaml
user_template: |-
{# some jinja template #}
```
2. template with custom functions
```yaml
user_template:
content: |-
{# some jinja template #}
custom_functions:
function_name: module_name:function_name
```
If you want to use custom functions in the template, you need to provide the `custom_functions` key, which is a dictionary of custom function names and their corresponding module paths in dot notation format.
### The Agent config
The `agents` section is a dictionary of agents, where the key is the agent's name and the value is the agent's config.
The agents need to act as a character in the scenes and respond to others' messages. So, the first part of the agent config is the backend config, which is used to communicate with the language model or user. You can use the `backend` key to specify the backend type and its config.
```yaml
agents:
Mike:
backend:
# user as the backend (a.k.a human-agent)
type: user
John:
backend:
# openai api as the backend
type: openai
model: gpt-3.5-turbo
temperature: 0.5
api_key:
base_url:
max_retries: 2
tool_choice:
type: auto
prop_validation_error_template: |-
{# some jinja template #}
```
You can also customize the backend by providing a object path of the custom backend class that implements the `Backend` abstract class.:
```yaml
agents:
Mike:
backend:
type: custom
path: module_name:CustomBackend
custom_config: value
```
```python
# module_name.py
from typing import Self
from operagents.prop import Prop
from operagents.timeline import Timeline
from operagents.config import CustomBackendConfig
from operagents.backend import Backend, Message, GenerateResponse, GeneratePropUsage
class CustomBackend(Backend):
@classmethod
def from_config(cls, config: CustomBackendConfig) -> Self:
return cls()
@overload
async def generate(
self,
timeline: Timeline,
messages: list[Message],
props: None = None,
) -> AsyncGenerator[GenerateResponse, None]: ...
@overload
async def generate(
self,
timeline: Timeline,
messages: list[Message],
props: list[Prop],
) -> AsyncGenerator[GenerateResponse | GeneratePropUsage, None]: ...
async def generate(
self, timeline: Timeline, messages: list[Message], props: list[Prop] | None = None
) -> AsyncGenerator[GenerateResponse | GeneratePropUsage, None]:
yield GenerateResponse(content="")
```
The next part of the agent config is the system/user template used to generate the context input for the language model. You can use the `system_template`/`user_template` key to specify the system/user template. Here is an example of the template config:
```yaml
agents:
John:
system_template: |-
Your name is {{ agent.name }}.
Current scene is {{ timeline.current_scene.name }}.
{% if timeline.current_scene.description -%}
{{ timeline.current_scene.description }}
{%- endif -%}
You are acting as {{ timeline.current_character.name }}.
{% if timeline.current_character.description -%}
{{ timeline.current_character.description }}
{%- endif -%}
Please continue the conversation on behalf of {{ agent.name }}({{ timeline.current_character.name }}) based on your known information and make your answer appear as natural and coherent as possible.
Please answer directly what you want to say and keep your reply as concise as possible.
user_template: |-
{% for event in timeline.past_events(agent) -%}
{% if event.type_ == "session_act" -%}
{{ event.character.agent_name }}({{ event.character.name }}): {{ event.content }}
{%- endif %}
{%- endfor %}
```
Another part of the agent config is the session summary system/user template, which is used to generate the summary of the scene session. You can use the `session_summary_system_template`/`session_summary_user_template` key to specify the session summary system/user template. Here is an example of the template config:
```yaml
agents:
John:
session_summary_system_template: |-
Your name is {{ agent.name }}.
Your task is to summarize the historical dialogue records according to the current scene, and summarize the most important information.
session_summary_user_template: |-
{% for event in agent.memory.get_memory_for_session(session_id) -%}
{% if event.type_ == "observe" -%}
{{ event.content }}
{%- elif event.type_ == "act" -%}
{{ agent.name }}({{ event.character.name }}): {{ event.content }}
{%- endif %}
{%- endfor %}
{% for event in timeline.session_past_events(agent, session_id) -%}
{% if event.type_ == "session_act" -%}
{{ event.character.agent_name }}({{ event.character.name }}): {{ event.content }}
{%- endif %}
{%- endfor %}
```
### Opening scene config
The `opening_scene` key is used to specify the start scene of the opera. The value is the name of the opening scene.
```yaml
opening_scene: "Introduction"
```
### The Scene config
The `scenes` section is a dictionary of scenes, where the key is the scene's name and the value is the scene's config.
The opera is composed of multiple scenes, and each scene has a number of characters. You first need to define the name, description (optional), and characters of the scene.
```yaml
scenes:
talking:
description: "The scene is about two people talking."
characters:
user:
agent_name: "Mike"
ai assistant:
agent_name: "John"
description: |-
You are a helpful assistant.
props: []
```
The characters in the scene must define the `agent_name` key, which is the name of the agent acting as the character. The `description` key (optional) can be used to describe the character in the agent template. The `props` key (optional) can be used to define the props of the character, see [the Prop config](#the-prop-config) for more details.
The `Flow` of the scene is designed to control the order of the characters' acting. You can specify the type and the parameters of the `Flow`.
1. `order` type
The `order` type is used to pre-define the order of the characters' acting. The characters will cycle through the order list until the scene ends.
```yaml
scenes:
talking:
flow:
type: order
order:
- user
- ai assistant
```
2. `model` type
The `model` type is used to specify the model to predict the next character to act. The model will predict the next character based on the current context.
```yaml
scenes:
talking:
flow:
type: model
backend:
type: openai
model: gpt-3.5-turbo
temperature: 0.5
system_template: ""
user_template: ""
allowed_characters: # optional, the characters allowed to act
- user
- ai assistant
begin_character: user # optional, the first character to act
fallback_character: ai assistant # optional, the fallback character when the model fails to predict
```
3. `user` type
The `user` type allows human to choose the next character to act.
```yaml
scenes:
talking:
flow:
type: user
```
4. `custom` type
The `custom` type allows you to define a custom flow class to control the order of the characters' acting.
```yaml
scenes:
talking:
flow:
type: custom
path: module_name:CustomFlow
custom_config: value
```
```python
# module_name.py
from typing import Self
from operagents.flow import Flow
from operagents.timeline import Timeline
from operagents.character import Character
from operagents.config import CustomFlowConfig
class CustomFlow(Flow):
@classmethod
def from_config(cls, config: CustomFlowConfig) -> Self:
return cls()
async def begin(self, timeline: Timeline) -> Character:
return ""
async def next(self, timeline: Timeline) -> Character:
return ""
```
The `Director` of the scene is used to control the next scene to play. You can specify the type and the parameters of the Director.
1. `model` type
The `model` type is used to specify the model to predict the next scene to play. If no finish flag found or no scene name found, the curent scene will continue to play.
```yaml
scenes:
talking:
director:
type: model
backend:
type: openai
model: gpt-3.5-turbo
temperature: 0.5
system_template: ""
user_template: ""
allowed_scenes: # optional, the next scenes allowed to play
- walking
- running
finish_flag: "finish" # optional, the finish flag to end the opera
```
2. `user` type
The `user` type allows human to choose the next scene to play.
```yaml
scenes:
talking:
director:
type: user
```
3. `never` type
The `never` Director never ends the current scene. Useful when there is a single scene and you want to end the opera by a `Prop`.
```yaml
scenes:
talking:
director:
type: never
```
4. `custom` type
The `custom` type allows you to define a custom director class to control the next scene to play.
```yaml
scenes:
talking:
director:
type: custom
path: module_name:CustomDirector
custom_config: value
```
```python
# module_name.py
from typing import Self
from operagents.scene import Scene
from operagents.director import Director
from operagents.timeline import Timeline
from operagents.config import CustomDirectorConfig
class CustomDirector(Director):
@classmethod
def from_config(cls, config: CustomDirectorConfig) -> Self:
return cls()
async def next_scene(self, timeline: Timeline) -> Scene | None:
return None
```
The `prepare` section of the scene is used to defined the preparation steps before the scene starts. You can do some initialization work here.
1. `preface` type
You can make the character say something before the scene starts.
```yaml
scenes:
talking:
prepare:
- type: preface
character_name: ai assistant
content: |-
Hello, I am John, your AI assistant. How can I help you today?
```
2. `function` type
The `function` type will call the custom function before the scene starts.
```yaml
scenes:
talking:
prepare:
- type: function
function: module_name:function_name
```
The custom function will receive one parameter of type `operagents.timeline.Timeline`.
```python
# module_name.py
from operagents.timeline import Timeline
async def function_name(timeline: Timeline) -> None:
pass
```
3. `custom` type
The `custom` type will call the custom prepare class before the scene starts.
```yaml
scenes:
talking:
prepare:
- type: custom
path: module_name:CustomPrepare
custom_config: value
```
```python
# module_name.py
from typing import Self
from operagents.timeline import Timeline
from operagents.scene.prepare import ScenePrepare
from operagents.config import CustomScenePrepareConfig
class CustomScenePrepare(ScenePrepare):
@classmethod
def from_config(cls, config: CustomScenePrepareConfig) -> Self:
return cls()
async def prepare(self, timeline: Timeline) -> None:
pass
```
### The Prop config
The characters in the scene can use props to improve there acting. The `props` section is a list of props, where each prop is a dictionary with the prop type and the prop config.
1. `function` Prop
The `function` prop will call the custom function when the prop is used.
```yaml
scenes:
talking:
characters:
ai assistant:
props:
- type: function
function: module_name:function_name
exception_template: |-
{# some jinja template #}
```
The custom function should has no arguments or one argument of type `pydantic.BaseModel`.
```python
from pydantic import Field, BaseModel
from datetime import datetime, timezone
async def current_time() -> str:
"""Get the current real world time."""
return datetime.now(timezone.utc).astimezone().isoformat()
class Args(BaseModel):
name: str = Field(description="The name")
async def greet(args: Args) -> str:
"""Greet the name."""
return f"Hello, {args.name}!"
```
Note that the function's name and docstring will be used as the prop's name and description. You can also provide the description of the args by pydantic's `Field`. The exception template will be used to render response when the function raises an error.
2. `custom` Prop
The `custom` prop will call the custom prop class when the prop is used.
```yaml
scenes:
talking:
characters:
ai assistant:
props:
- type: custom
path: module_name:CustomProp
custom_config: value
```
```python
# module_name.py
from typing import Any, Self
from pydantic import BaseModel
from operagents.prop import Prop
from operagents.config import CustomPropConfig
class CustomProp(Prop):
"""The description of the prop"""
params: BaseModel | None
"""The parameters of the prop"""
@classmethod
def from_config(cls, config: CustomPropConfig) -> Self:
return cls()
async def call(self, params: BaseModel | None) -> Any:
return ""
```
### The Hook config
Hooks enables you to run custom code when specific timeline events occur. The `hooks` section is a list of hooks, where each hook is a dictionary with the hook type and the hook config. By default, operagents enables the `summary` hook unless you change the `hooks` section.
1. `summary` Hook
The `summary` hook will call the agents to summarize the session when the session ends. You can optionally specify the agent names to summarize.
```yaml
hooks:
- type: summary
agent_names:
- Mike
- John
```
2. `custom` Hook
The `custom` hook will invoke the custom hook class when specific timeline event encounters.
```yaml
hooks:
- type: custom
path: module_name:CustomHook
custom_config: value
```
```python
# module_name.py
from typing import Self
from operagents.hook import Hook
from operagents.timeline import Timeline
from operagents.config import CustomHookConfig
from operagents.timeline.event import (
TimelineEventEnd,
TimelineEventStart,
TimelineEventSessionAct,
TimelineEventSessionEnd,
TimelineEventSessionStart,
)
class CustomHook(Hook):
@classmethod
def from_config(cls, config: CustomHookConfig) -> Self:
return cls()
async def on_timeline_start(
self, timeline: Timeline, event: TimelineEventStart
):
"""Called when the timeline is started."""
pass
async def on_timeline_end(
self, timeline: Timeline, event: TimelineEventEnd
):
"""Called when the timeline is ended."""
pass
async def on_timeline_session_start(
self, timeline: Timeline, event: TimelineEventSessionStart
):
"""Called when a session is started."""
pass
async def on_timeline_session_end(
self, timeline: Timeline, event: TimelineEventSessionEnd
):
"""Called when a session is ended."""
pass
async def on_timeline_session_act(
self, timeline: Timeline, event: TimelineEventSessionAct
):
"""Called when a character acts in a session."""
pass
```
The hook class may contains methods in the format of `on_timeline_<event_type>`, where `<event_type>` is the type of the timeline event.
### Run the opera
operagents provides a command-line tool to easily run the opera. You can run the opera with the following command:
```bash
operagents run config.yaml
```
If you want to see the debug logs, you can set the `--log-level` option:
```bash
operagents run --log-level DEBUG config.yaml
```
More commands and options can be found by running `operagents --help`.
If you want to run the opera programmatically, you can use the `opera.run` function:
```python
import asyncio
from pathlib import Path
import yaml
from operagents.opera import Opera
from operagents.log import setup_logging
from operagents.config import OperagentsConfig
async def main():
# if you want to setup the default logging for operagents
setup_logging("INFO")
# load the opera from config file
opera = Opera.from_config(
OperagentsConfig.model_validate(
yaml.safe_load(Path("./config.yaml").read_text(encoding="utf-8"))
)
)
finish_state = await opera.run()
if __name__ == "__main__":
asyncio.run(main())
```
## Examples
### Chatbot
```bash
cd examples/chatbot
env OPENAI_API_KEY=sk-xxx OPENAI_BASE_URL=https://api.openai.com/v1 operagents run --log-level DEBUG config.yaml
```
## Development
Open in Codespaces (Dev Container):
[](https://github.com/codespaces/new?hide_repo_select=true&ref=master&repo=767939984)
Or install the development environment locally with:
```bash
poetry install && poetry run pre-commit install
```
Raw data
{
"_id": null,
"home_page": null,
"name": "operagents",
"maintainer": null,
"docs_url": null,
"requires_python": "<4.0,>=3.10",
"maintainer_email": null,
"keywords": "agent, llm, langchain, sop, langgraph",
"author": "yanyongyu",
"author_email": "yyy@yyydl.top",
"download_url": "https://files.pythonhosted.org/packages/2e/e2/27161c9b67fe8190355849ff5ca4684061c0cb163b2442fa50079731ab02/operagents-0.0.15.tar.gz",
"platform": null,
"description": "# Operagents\n\n## Installation\n\nInstall the latest version with:\n\n```bash\npip install operagents\n# or use poetry\npoetry add operagents\n# or use pdm\npdm add operagents\n# or use uv\nuv pip install operagents\n```\n\n## Concepts\n\n### Agent\n\nAn agent is a human or a language model that can act as [characters](#character) and use [props](#prop) in the opera scenes. The agent can communicate with others by observing and acting. Every agent has a backend (e.g. user, openai api) to generate the response and own memory to store the long-term / short-term information.\n\n### Scene\n\nA scene is a part of the opera that contains a number of [characters](#character). Every scene has a [flow](#flow) and a [director](#director) to control the whole [session](#session) process. The scene can also have a prepare section to do some initialization work before the scene starts.\n\n### Character\n\nA character is a role in the [scene](#scene). Every character has a name, a description, and a list of [props](#prop). When the scene starts, an agent will act as the character and communicate with others.\n\n### Flow\n\nThe flow is used to control the order of the characters' acting in the [scene](#scene).\n\n### Director\n\nThe director is used to decide whether to end the current scene and which scene to play next.\n\n### Prop\n\nA prop is a tool that can be used by the [agents](#agent) to improve their acting. Agents can get external information by using props.\n\n### Timeline\n\nThe timeline is the main runtime component of the opera to manage the [session](#session) process. It runs the current session and switches between sessions. The timeline also records the global information of the opera, and can be shared by all agents.\n\n### Session\n\nA session indicates a single run of the scene. It contains an unique identifier and its corresponding scene.\n\n## Usage\n\nThe common way to use operagents is to write a config file and run the opera with the `operagents` command-line tool.\n\n### Start writing a config file\n\nCreate a `config.yaml` file with the following basic content:\n\n```yaml\n# yaml-language-server: $schema=https://operagents.yyydl.top/schemas/config.schema.json\n\nagents:\nopening_scene: \"\"\nscenes:\n```\n\nThe first line is a comment that tells the YAML Language Server to use the schema from the specified URL. This will enable autocompletion and validation in your editor. \nThe schema is related to the version of the operagents framework you are using. The URL is in the format `https://operagents.yyydl.top/schemas/config-<version>.schema.json`, where `<version>` is the version of the framework, e.g. `0.0.1`. If no version is specified, the latest (master) version will be used.\n\n### The Template config\n\nBefore writing the agent and scene configs, we need to learn about the template config.\n\nOperagents uses templates to generate the context input for the language model. A template is a string in [jinja](https://jinja.palletsprojects.com/) format. You can use jinja2 syntax with provided context varaibles to control the input to the language model.\n\nA template config can be in the following format:\n\n1. simple string template\n\n ```yaml\n user_template: |-\n {# some jinja template #}\n ```\n\n2. template with custom functions\n\n ```yaml\n user_template:\n content: |-\n {# some jinja template #}\n custom_functions:\n function_name: module_name:function_name\n ```\n\nIf you want to use custom functions in the template, you need to provide the `custom_functions` key, which is a dictionary of custom function names and their corresponding module paths in dot notation format.\n\n### The Agent config\n\nThe `agents` section is a dictionary of agents, where the key is the agent's name and the value is the agent's config.\n\nThe agents need to act as a character in the scenes and respond to others' messages. So, the first part of the agent config is the backend config, which is used to communicate with the language model or user. You can use the `backend` key to specify the backend type and its config.\n\n```yaml\nagents:\n Mike:\n backend:\n # user as the backend (a.k.a human-agent)\n type: user\n John:\n backend:\n # openai api as the backend\n type: openai\n model: gpt-3.5-turbo\n temperature: 0.5\n api_key:\n base_url:\n max_retries: 2\n tool_choice:\n type: auto\n prop_validation_error_template: |-\n {# some jinja template #}\n```\n\nYou can also customize the backend by providing a object path of the custom backend class that implements the `Backend` abstract class.:\n\n```yaml\nagents:\n Mike:\n backend:\n type: custom\n path: module_name:CustomBackend\n custom_config: value\n```\n\n```python\n# module_name.py\n\nfrom typing import Self\n\nfrom operagents.prop import Prop\nfrom operagents.timeline import Timeline\nfrom operagents.config import CustomBackendConfig\nfrom operagents.backend import Backend, Message, GenerateResponse, GeneratePropUsage\n\n\nclass CustomBackend(Backend):\n @classmethod\n def from_config(cls, config: CustomBackendConfig) -> Self:\n return cls()\n\n @overload\n async def generate(\n self,\n timeline: Timeline,\n messages: list[Message],\n props: None = None,\n ) -> AsyncGenerator[GenerateResponse, None]: ...\n\n @overload\n async def generate(\n self,\n timeline: Timeline,\n messages: list[Message],\n props: list[Prop],\n ) -> AsyncGenerator[GenerateResponse | GeneratePropUsage, None]: ...\n\n async def generate(\n self, timeline: Timeline, messages: list[Message], props: list[Prop] | None = None\n ) -> AsyncGenerator[GenerateResponse | GeneratePropUsage, None]:\n yield GenerateResponse(content=\"\")\n```\n\nThe next part of the agent config is the system/user template used to generate the context input for the language model. You can use the `system_template`/`user_template` key to specify the system/user template. Here is an example of the template config:\n\n```yaml\nagents:\n John:\n system_template: |-\n Your name is {{ agent.name }}.\n Current scene is {{ timeline.current_scene.name }}.\n {% if timeline.current_scene.description -%}\n {{ timeline.current_scene.description }}\n {%- endif -%}\n You are acting as {{ timeline.current_character.name }}.\n {% if timeline.current_character.description -%}\n {{ timeline.current_character.description }}\n {%- endif -%}\n Please continue the conversation on behalf of {{ agent.name }}({{ timeline.current_character.name }}) based on your known information and make your answer appear as natural and coherent as possible.\n Please answer directly what you want to say and keep your reply as concise as possible.\n user_template: |-\n {% for event in timeline.past_events(agent) -%}\n {% if event.type_ == \"session_act\" -%}\n {{ event.character.agent_name }}({{ event.character.name }}): {{ event.content }}\n {%- endif %}\n {%- endfor %}\n```\n\nAnother part of the agent config is the session summary system/user template, which is used to generate the summary of the scene session. You can use the `session_summary_system_template`/`session_summary_user_template` key to specify the session summary system/user template. Here is an example of the template config:\n\n```yaml\nagents:\n John:\n session_summary_system_template: |-\n Your name is {{ agent.name }}.\n Your task is to summarize the historical dialogue records according to the current scene, and summarize the most important information.\n session_summary_user_template: |-\n {% for event in agent.memory.get_memory_for_session(session_id) -%}\n {% if event.type_ == \"observe\" -%}\n {{ event.content }}\n {%- elif event.type_ == \"act\" -%}\n {{ agent.name }}({{ event.character.name }}): {{ event.content }}\n {%- endif %}\n {%- endfor %}\n {% for event in timeline.session_past_events(agent, session_id) -%}\n {% if event.type_ == \"session_act\" -%}\n {{ event.character.agent_name }}({{ event.character.name }}): {{ event.content }}\n {%- endif %}\n {%- endfor %}\n```\n\n### Opening scene config\n\nThe `opening_scene` key is used to specify the start scene of the opera. The value is the name of the opening scene.\n\n```yaml\nopening_scene: \"Introduction\"\n```\n\n### The Scene config\n\nThe `scenes` section is a dictionary of scenes, where the key is the scene's name and the value is the scene's config.\n\nThe opera is composed of multiple scenes, and each scene has a number of characters. You first need to define the name, description (optional), and characters of the scene.\n\n```yaml\nscenes:\n talking:\n description: \"The scene is about two people talking.\"\n characters:\n user:\n agent_name: \"Mike\"\n ai assistant:\n agent_name: \"John\"\n description: |-\n You are a helpful assistant.\n props: []\n```\n\nThe characters in the scene must define the `agent_name` key, which is the name of the agent acting as the character. The `description` key (optional) can be used to describe the character in the agent template. The `props` key (optional) can be used to define the props of the character, see [the Prop config](#the-prop-config) for more details.\n\nThe `Flow` of the scene is designed to control the order of the characters' acting. You can specify the type and the parameters of the `Flow`.\n\n1. `order` type\n\n The `order` type is used to pre-define the order of the characters' acting. The characters will cycle through the order list until the scene ends.\n\n ```yaml\n scenes:\n talking:\n flow:\n type: order\n order:\n - user\n - ai assistant\n ```\n\n2. `model` type\n\n The `model` type is used to specify the model to predict the next character to act. The model will predict the next character based on the current context.\n\n ```yaml\n scenes:\n talking:\n flow:\n type: model\n backend:\n type: openai\n model: gpt-3.5-turbo\n temperature: 0.5\n system_template: \"\"\n user_template: \"\"\n allowed_characters: # optional, the characters allowed to act\n - user\n - ai assistant\n begin_character: user # optional, the first character to act\n fallback_character: ai assistant # optional, the fallback character when the model fails to predict\n ```\n\n3. `user` type\n\n The `user` type allows human to choose the next character to act.\n\n ```yaml\n scenes:\n talking:\n flow:\n type: user\n ```\n\n4. `custom` type\n\n The `custom` type allows you to define a custom flow class to control the order of the characters' acting.\n\n ```yaml\n scenes:\n talking:\n flow:\n type: custom\n path: module_name:CustomFlow\n custom_config: value\n ```\n\n ```python\n # module_name.py\n\n from typing import Self\n\n from operagents.flow import Flow\n from operagents.timeline import Timeline\n from operagents.character import Character\n from operagents.config import CustomFlowConfig\n\n\n class CustomFlow(Flow):\n @classmethod\n def from_config(cls, config: CustomFlowConfig) -> Self:\n return cls()\n\n async def begin(self, timeline: Timeline) -> Character:\n return \"\"\n\n async def next(self, timeline: Timeline) -> Character:\n return \"\"\n ```\n\nThe `Director` of the scene is used to control the next scene to play. You can specify the type and the parameters of the Director.\n\n1. `model` type\n\n The `model` type is used to specify the model to predict the next scene to play. If no finish flag found or no scene name found, the curent scene will continue to play.\n\n ```yaml\n scenes:\n talking:\n director:\n type: model\n backend:\n type: openai\n model: gpt-3.5-turbo\n temperature: 0.5\n system_template: \"\"\n user_template: \"\"\n allowed_scenes: # optional, the next scenes allowed to play\n - walking\n - running\n finish_flag: \"finish\" # optional, the finish flag to end the opera\n ```\n\n2. `user` type\n\n The `user` type allows human to choose the next scene to play.\n\n ```yaml\n scenes:\n talking:\n director:\n type: user\n ```\n\n3. `never` type\n\n The `never` Director never ends the current scene. Useful when there is a single scene and you want to end the opera by a `Prop`.\n\n ```yaml\n scenes:\n talking:\n director:\n type: never\n ```\n\n4. `custom` type\n\n The `custom` type allows you to define a custom director class to control the next scene to play.\n\n ```yaml\n scenes:\n talking:\n director:\n type: custom\n path: module_name:CustomDirector\n custom_config: value\n ```\n\n ```python\n # module_name.py\n\n from typing import Self\n\n from operagents.scene import Scene\n from operagents.director import Director\n from operagents.timeline import Timeline\n from operagents.config import CustomDirectorConfig\n\n class CustomDirector(Director):\n @classmethod\n def from_config(cls, config: CustomDirectorConfig) -> Self:\n return cls()\n\n async def next_scene(self, timeline: Timeline) -> Scene | None:\n return None\n ```\n\nThe `prepare` section of the scene is used to defined the preparation steps before the scene starts. You can do some initialization work here.\n\n1. `preface` type\n\n You can make the character say something before the scene starts.\n\n ```yaml\n scenes:\n talking:\n prepare:\n - type: preface\n character_name: ai assistant\n content: |-\n Hello, I am John, your AI assistant. How can I help you today?\n ```\n\n2. `function` type\n\n The `function` type will call the custom function before the scene starts.\n\n ```yaml\n scenes:\n talking:\n prepare:\n - type: function\n function: module_name:function_name\n ```\n\n The custom function will receive one parameter of type `operagents.timeline.Timeline`.\n\n ```python\n # module_name.py\n\n from operagents.timeline import Timeline\n\n async def function_name(timeline: Timeline) -> None:\n pass\n ```\n\n3. `custom` type\n\n The `custom` type will call the custom prepare class before the scene starts.\n\n ```yaml\n scenes:\n talking:\n prepare:\n - type: custom\n path: module_name:CustomPrepare\n custom_config: value\n ```\n\n ```python\n # module_name.py\n\n from typing import Self\n\n from operagents.timeline import Timeline\n from operagents.scene.prepare import ScenePrepare\n from operagents.config import CustomScenePrepareConfig\n\n class CustomScenePrepare(ScenePrepare):\n @classmethod\n def from_config(cls, config: CustomScenePrepareConfig) -> Self:\n return cls()\n\n async def prepare(self, timeline: Timeline) -> None:\n pass\n ```\n\n### The Prop config\n\nThe characters in the scene can use props to improve there acting. The `props` section is a list of props, where each prop is a dictionary with the prop type and the prop config.\n\n1. `function` Prop\n\n The `function` prop will call the custom function when the prop is used.\n\n ```yaml\n scenes:\n talking:\n characters:\n ai assistant:\n props:\n - type: function\n function: module_name:function_name\n exception_template: |-\n {# some jinja template #}\n ```\n\n The custom function should has no arguments or one argument of type `pydantic.BaseModel`.\n\n ```python\n from pydantic import Field, BaseModel\n from datetime import datetime, timezone\n\n async def current_time() -> str:\n \"\"\"Get the current real world time.\"\"\"\n return datetime.now(timezone.utc).astimezone().isoformat()\n\n class Args(BaseModel):\n name: str = Field(description=\"The name\")\n\n async def greet(args: Args) -> str:\n \"\"\"Greet the name.\"\"\"\n return f\"Hello, {args.name}!\"\n ```\n\n Note that the function's name and docstring will be used as the prop's name and description. You can also provide the description of the args by pydantic's `Field`. The exception template will be used to render response when the function raises an error.\n\n2. `custom` Prop\n\n The `custom` prop will call the custom prop class when the prop is used.\n\n ```yaml\n scenes:\n talking:\n characters:\n ai assistant:\n props:\n - type: custom\n path: module_name:CustomProp\n custom_config: value\n ```\n\n ```python\n # module_name.py\n\n from typing import Any, Self\n\n from pydantic import BaseModel\n from operagents.prop import Prop\n from operagents.config import CustomPropConfig\n\n class CustomProp(Prop):\n \"\"\"The description of the prop\"\"\"\n\n params: BaseModel | None\n \"\"\"The parameters of the prop\"\"\"\n\n @classmethod\n def from_config(cls, config: CustomPropConfig) -> Self:\n return cls()\n\n async def call(self, params: BaseModel | None) -> Any:\n return \"\"\n ```\n\n### The Hook config\n\nHooks enables you to run custom code when specific timeline events occur. The `hooks` section is a list of hooks, where each hook is a dictionary with the hook type and the hook config. By default, operagents enables the `summary` hook unless you change the `hooks` section.\n\n1. `summary` Hook\n\n The `summary` hook will call the agents to summarize the session when the session ends. You can optionally specify the agent names to summarize.\n\n ```yaml\n hooks:\n - type: summary\n agent_names:\n - Mike\n - John\n ```\n\n2. `custom` Hook\n\n The `custom` hook will invoke the custom hook class when specific timeline event encounters.\n\n ```yaml\n hooks:\n - type: custom\n path: module_name:CustomHook\n custom_config: value\n ```\n\n ```python\n # module_name.py\n\n from typing import Self\n\n from operagents.hook import Hook\n from operagents.timeline import Timeline\n from operagents.config import CustomHookConfig\n from operagents.timeline.event import (\n TimelineEventEnd,\n TimelineEventStart,\n TimelineEventSessionAct,\n TimelineEventSessionEnd,\n TimelineEventSessionStart,\n )\n\n class CustomHook(Hook):\n @classmethod\n def from_config(cls, config: CustomHookConfig) -> Self:\n return cls()\n\n async def on_timeline_start(\n self, timeline: Timeline, event: TimelineEventStart\n ):\n \"\"\"Called when the timeline is started.\"\"\"\n pass\n\n async def on_timeline_end(\n self, timeline: Timeline, event: TimelineEventEnd\n ):\n \"\"\"Called when the timeline is ended.\"\"\"\n pass\n\n async def on_timeline_session_start(\n self, timeline: Timeline, event: TimelineEventSessionStart\n ):\n \"\"\"Called when a session is started.\"\"\"\n pass\n\n async def on_timeline_session_end(\n self, timeline: Timeline, event: TimelineEventSessionEnd\n ):\n \"\"\"Called when a session is ended.\"\"\"\n pass\n\n async def on_timeline_session_act(\n self, timeline: Timeline, event: TimelineEventSessionAct\n ):\n \"\"\"Called when a character acts in a session.\"\"\"\n pass\n ```\n\n The hook class may contains methods in the format of `on_timeline_<event_type>`, where `<event_type>` is the type of the timeline event.\n\n### Run the opera\n\noperagents provides a command-line tool to easily run the opera. You can run the opera with the following command:\n\n```bash\noperagents run config.yaml\n```\n\nIf you want to see the debug logs, you can set the `--log-level` option:\n\n```bash\noperagents run --log-level DEBUG config.yaml\n```\n\nMore commands and options can be found by running `operagents --help`.\n\nIf you want to run the opera programmatically, you can use the `opera.run` function:\n\n```python\nimport asyncio\nfrom pathlib import Path\n\nimport yaml\nfrom operagents.opera import Opera\nfrom operagents.log import setup_logging\nfrom operagents.config import OperagentsConfig\n\n\nasync def main():\n # if you want to setup the default logging for operagents\n setup_logging(\"INFO\")\n\n # load the opera from config file\n opera = Opera.from_config(\n OperagentsConfig.model_validate(\n yaml.safe_load(Path(\"./config.yaml\").read_text(encoding=\"utf-8\"))\n )\n )\n\n finish_state = await opera.run()\n\n\nif __name__ == \"__main__\":\n asyncio.run(main())\n```\n\n## Examples\n\n### Chatbot\n\n```bash\ncd examples/chatbot\nenv OPENAI_API_KEY=sk-xxx OPENAI_BASE_URL=https://api.openai.com/v1 operagents run --log-level DEBUG config.yaml\n```\n\n## Development\n\nOpen in Codespaces (Dev Container):\n\n[](https://github.com/codespaces/new?hide_repo_select=true&ref=master&repo=767939984)\n\nOr install the development environment locally with:\n\n```bash\npoetry install && poetry run pre-commit install\n```\n\n",
"bugtrack_url": null,
"license": "MIT",
"summary": "Dynamic, highly customizable language agents framework",
"version": "0.0.15",
"project_urls": null,
"split_keywords": [
"agent",
" llm",
" langchain",
" sop",
" langgraph"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "8e9749044e5dc0412eeea4e00cd91c696df501e4cfe19acb1d52ad81bfe9878b",
"md5": "b3ea0b16be9d43cb898224d6816e6f29",
"sha256": "8ef13210cc1face32dfc5fbdea983c64bf1be1c340dfaf81ac7d90eb9c6e08ae"
},
"downloads": -1,
"filename": "operagents-0.0.15-py3-none-any.whl",
"has_sig": false,
"md5_digest": "b3ea0b16be9d43cb898224d6816e6f29",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": "<4.0,>=3.10",
"size": 46067,
"upload_time": "2024-05-11T10:15:12",
"upload_time_iso_8601": "2024-05-11T10:15:12.675516Z",
"url": "https://files.pythonhosted.org/packages/8e/97/49044e5dc0412eeea4e00cd91c696df501e4cfe19acb1d52ad81bfe9878b/operagents-0.0.15-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "2ee227161c9b67fe8190355849ff5ca4684061c0cb163b2442fa50079731ab02",
"md5": "82ff53531673acc8a16e0eedb3ee955a",
"sha256": "51bd7d3bd80f77de0c0bcd93ecd085ae1188630a44117fdca6453cceace282f4"
},
"downloads": -1,
"filename": "operagents-0.0.15.tar.gz",
"has_sig": false,
"md5_digest": "82ff53531673acc8a16e0eedb3ee955a",
"packagetype": "sdist",
"python_version": "source",
"requires_python": "<4.0,>=3.10",
"size": 35825,
"upload_time": "2024-05-11T10:15:14",
"upload_time_iso_8601": "2024-05-11T10:15:14.009177Z",
"url": "https://files.pythonhosted.org/packages/2e/e2/27161c9b67fe8190355849ff5ca4684061c0cb163b2442fa50079731ab02/operagents-0.0.15.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-05-11 10:15:14",
"github": false,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"lcname": "operagents"
}
